fs/gfs2/log.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 #include <linux/sched.h>
   8 #include <linux/slab.h>
   9 #include <linux/spinlock.h>
  10 #include <linux/completion.h>
  11 #include <linux/buffer_head.h>
  12 #include <linux/gfs2_ondisk.h>
  13 #include <linux/crc32.h>
  14 #include <linux/crc32c.h>
  15 #include <linux/delay.h>
  16 #include <linux/kthread.h>
  17 #include <linux/freezer.h>
  18 #include <linux/bio.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/writeback.h>
  21 #include <linux/list_sort.h>
  22
  23 #include "gfs2.h"
  24 #include "incore.h"
  25 #include "bmap.h"
  26 #include "glock.h"
  27 #include "log.h"
  28 #include "lops.h"
  29 #include "meta_io.h"
  30 #include "util.h"
  31 #include "dir.h"
  32 #include "trace_gfs2.h"
  33 #include "trans.h"
  34
  35 static void gfs2_log_shutdown(struct gfs2_sbd *sdp);
  36
  37 /**
  38  * gfs2_struct2blk - compute stuff
  39  * @sdp: the filesystem
  40  * @nstruct: the number of structures
  41  *
  42  * Compute the number of log descriptor blocks needed to hold a certain number
  43  * of structures of a certain size.
  44  *
  45  * Returns: the number of blocks needed (minimum is always 1)
  46  */
  47
  48 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct)
  49 {
  50         unsigned int blks;
  51         unsigned int first, second;
  52
  53         blks = 1;
  54         first = sdp->sd_ldptrs;
  55
  56         if (nstruct > first) {
  57                 second = sdp->sd_inptrs;
  58                 blks += DIV_ROUND_UP(nstruct - first, second);
  59         }
  60
  61         return blks;
  62 }
  63
  64 /**
  65  * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
  66  * @mapping: The associated mapping (maybe NULL)
  67  * @bd: The gfs2_bufdata to remove
  68  *
  69  * The ail lock _must_ be held when calling this function
  70  *
  71  */
  72
  73 void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
  74 {
  75         bd->bd_tr = NULL;
  76         list_del_init(&bd->bd_ail_st_list);
  77         list_del_init(&bd->bd_ail_gl_list);
  78         atomic_dec(&bd->bd_gl->gl_ail_count);
  79         brelse(bd->bd_bh);
  80 }
  81
  82 /**
  83  * gfs2_ail1_start_one - Start I/O on a part of the AIL
  84  * @sdp: the filesystem
  85  * @wbc: The writeback control structure
  86  * @ai: The ail structure
  87  *
  88  */
  89
  90 static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
  91                                struct writeback_control *wbc,
  92                                struct gfs2_trans *tr)
  93 __releases(&sdp->sd_ail_lock)
  94 __acquires(&sdp->sd_ail_lock)
  95 {
  96         struct gfs2_glock *gl = NULL;
  97         struct address_space *mapping;
  98         struct gfs2_bufdata *bd, *s;
  99         struct buffer_head *bh;
 100         int ret = 0;
 101
 102         list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
 103                 bh = bd->bd_bh;
 104
 105                 gfs2_assert(sdp, bd->bd_tr == tr);
 106
 107                 if (!buffer_busy(bh)) {
 108                         if (buffer_uptodate(bh)) {
 109                                 list_move(&bd->bd_ail_st_list,
 110                                           &tr->tr_ail2_list);
 111                                 continue;
 112                         }
 113                         if (!cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
 114                                 gfs2_io_error_bh(sdp, bh);
 115                                 gfs2_withdraw_delayed(sdp);
 116                         }
 117                 }
 118
 119                 if (gfs2_withdrawn(sdp)) {
 120                         gfs2_remove_from_ail(bd);
 121                         continue;
 122                 }
 123                 if (!buffer_dirty(bh))
 124                         continue;
 125                 if (gl == bd->bd_gl)
 126                         continue;
 127                 gl = bd->bd_gl;
 128                 list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
 129                 mapping = bh->b_page->mapping;
 130                 if (!mapping)
 131                         continue;
 132                 spin_unlock(&sdp->sd_ail_lock);
 133                 ret = generic_writepages(mapping, wbc);
 134                 spin_lock(&sdp->sd_ail_lock);
 135                 if (ret || wbc->nr_to_write <= 0)
 136                         break;
 137                 return -EBUSY;
 138         }
 139
 140         return ret;
 141 }
 142
 143 static void dump_ail_list(struct gfs2_sbd *sdp)
 144 {
 145         struct gfs2_trans *tr;
 146         struct gfs2_bufdata *bd;
 147         struct buffer_head *bh;
 148
 149         list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
 150                 list_for_each_entry_reverse(bd, &tr->tr_ail1_list,
 151                                             bd_ail_st_list) {
 152                         bh = bd->bd_bh;
 153                         fs_err(sdp, "bd %p: blk:0x%llx bh=%p ", bd,
 154                                (unsigned long long)bd->bd_blkno, bh);
 155                         if (!bh) {
 156                                 fs_err(sdp, "\n");
 157                                 continue;
 158                         }
 159                         fs_err(sdp, "0x%llx up2:%d dirt:%d lkd:%d req:%d "
 160                                "map:%d new:%d ar:%d aw:%d delay:%d "
 161                                "io err:%d unwritten:%d dfr:%d pin:%d esc:%d\n",
 162                                (unsigned long long)bh->b_blocknr,
 163                                buffer_uptodate(bh), buffer_dirty(bh),
 164                                buffer_locked(bh), buffer_req(bh),
 165                                buffer_mapped(bh), buffer_new(bh),
 166                                buffer_async_read(bh), buffer_async_write(bh),
 167                                buffer_delay(bh), buffer_write_io_error(bh),
 168                                buffer_unwritten(bh),
 169                                buffer_defer_completion(bh),
 170                                buffer_pinned(bh), buffer_escaped(bh));
 171                 }
 172         }
 173 }
 174
 175 /**
 176  * gfs2_ail1_flush - start writeback of some ail1 entries
 177  * @sdp: The super block
 178  * @wbc: The writeback control structure
 179  *
 180  * Writes back some ail1 entries, according to the limits in the
 181  * writeback control structure
 182  */
 183
 184 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
 185 {
 186         struct list_head *head = &sdp->sd_ail1_list;
 187         struct gfs2_trans *tr;
 188         struct blk_plug plug;
 189         int ret;
 190         unsigned long flush_start = jiffies;
 191
 192         trace_gfs2_ail_flush(sdp, wbc, 1);
 193         blk_start_plug(&plug);
 194         spin_lock(&sdp->sd_ail_lock);
 195 restart:
 196         ret = 0;
 197         if (time_after(jiffies, flush_start + (HZ * 600))) {
 198                 fs_err(sdp, "Error: In %s for ten minutes! t=%d\n",
 199                        __func__, current->journal_info ? 1 : 0);
 200                 dump_ail_list(sdp);
 201                 goto out;
 202         }
 203         list_for_each_entry_reverse(tr, head, tr_list) {
 204                 if (wbc->nr_to_write <= 0)
 205                         break;
 206                 ret = gfs2_ail1_start_one(sdp, wbc, tr);
 207                 if (ret) {
 208                         if (ret == -EBUSY)
 209                                 goto restart;
 210                         break;
 211                 }
 212         }
 213 out:
 214         spin_unlock(&sdp->sd_ail_lock);
 215         blk_finish_plug(&plug);
 216         if (ret) {
 217                 gfs2_lm(sdp, "gfs2_ail1_start_one (generic_writepages) "
 218                         "returned: %d\n", ret);
 219                 gfs2_withdraw(sdp);
 220         }
 221         trace_gfs2_ail_flush(sdp, wbc, 0);
 222 }
 223
 224 /**
 225  * gfs2_ail1_start - start writeback of all ail1 entries
 226  * @sdp: The superblock
 227  */
 228
 229 static void gfs2_ail1_start(struct gfs2_sbd *sdp)
 230 {
 231         struct writeback_control wbc = {
 232                 .sync_mode = WB_SYNC_NONE,
 233                 .nr_to_write = LONG_MAX,
 234                 .range_start = 0,
 235                 .range_end = LLONG_MAX,
 236         };
 237
 238         return gfs2_ail1_flush(sdp, &wbc);
 239 }
 240
 241 /**
 242  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 243  * @sdp: the filesystem
 244  * @tr: the transaction
 245  * @max_revokes: If nonzero, issue revokes for the bd items for written buffers
 246  *
 247  * returns: the transaction's count of remaining active items
 248  */
 249
 250 static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
 251                                 int *max_revokes)
 252 {
 253         struct gfs2_bufdata *bd, *s;
 254         struct buffer_head *bh;
 255         int active_count = 0;
 256
 257         list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
 258                                          bd_ail_st_list) {
 259                 bh = bd->bd_bh;
 260                 gfs2_assert(sdp, bd->bd_tr == tr);
 261                 /*
 262                  * If another process flagged an io error, e.g. writing to the
 263                  * journal, error all other bhs and move them off the ail1 to
 264                  * prevent a tight loop when unmount tries to flush ail1,
 265                  * regardless of whether they're still busy. If no outside
 266                  * errors were found and the buffer is busy, move to the next.
 267                  * If the ail buffer is not busy and caught an error, flag it
 268                  * for others.
 269                  */
 270                 if (!sdp->sd_log_error && buffer_busy(bh)) {
 271                         active_count++;
 272                         continue;
 273                 }
 274                 if (!buffer_uptodate(bh) &&
 275                     !cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
 276                         gfs2_io_error_bh(sdp, bh);
 277                         gfs2_withdraw_delayed(sdp);
 278                 }
 279                 /*
 280                  * If we have space for revokes and the bd is no longer on any
 281                  * buf list, we can just add a revoke for it immediately and
 282                  * avoid having to put it on the ail2 list, where it would need
 283                  * to be revoked later.
 284                  */
 285                 if (*max_revokes && list_empty(&bd->bd_list)) {
 286                         gfs2_add_revoke(sdp, bd);
 287                         (*max_revokes)--;
 288                         continue;
 289                 }
 290                 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
 291         }
 292         return active_count;
 293 }
 294
 295 /**
 296  * gfs2_ail1_empty - Try to empty the ail1 lists
 297  * @sdp: The superblock
 298  * @max_revokes: If non-zero, add revokes where appropriate
 299  *
 300  * Tries to empty the ail1 lists, starting with the oldest first
 301  */
 302
 303 static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int max_revokes)
 304 {
 305         struct gfs2_trans *tr, *s;
 306         int oldest_tr = 1;
 307         int ret;
 308
 309         spin_lock(&sdp->sd_ail_lock);
 310         list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
 311                 if (!gfs2_ail1_empty_one(sdp, tr, &max_revokes) && oldest_tr)
 312                         list_move(&tr->tr_list, &sdp->sd_ail2_list);
 313                 else
 314                         oldest_tr = 0;
 315         }
 316         ret = list_empty(&sdp->sd_ail1_list);
 317         spin_unlock(&sdp->sd_ail_lock);
 318
 319         if (test_bit(SDF_WITHDRAWING, &sdp->sd_flags)) {
 320                 gfs2_lm(sdp, "fatal: I/O error(s)\n");
 321                 gfs2_withdraw(sdp);
 322         }
 323
 324         return ret;
 325 }
 326
 327 static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
 328 {
 329         struct gfs2_trans *tr;
 330         struct gfs2_bufdata *bd;
 331         struct buffer_head *bh;
 332
 333         spin_lock(&sdp->sd_ail_lock);
 334         list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
 335                 list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
 336                         bh = bd->bd_bh;
 337                         if (!buffer_locked(bh))
 338                                 continue;
 339                         get_bh(bh);
 340                         spin_unlock(&sdp->sd_ail_lock);
 341                         wait_on_buffer(bh);
 342                         brelse(bh);
 343                         return;
 344                 }
 345         }
 346         spin_unlock(&sdp->sd_ail_lock);
 347 }
 348
 349 /**
 350  * gfs2_ail_empty_tr - empty one of the ail lists for a transaction
 351  */
 352
 353 static void gfs2_ail_empty_tr(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
 354                               struct list_head *head)
 355 {
 356         struct gfs2_bufdata *bd;
 357
 358         while (!list_empty(head)) {
 359                 bd = list_first_entry(head, struct gfs2_bufdata,
 360                                       bd_ail_st_list);
 361                 gfs2_assert(sdp, bd->bd_tr == tr);
 362                 gfs2_remove_from_ail(bd);
 363         }
 364 }
 365
 366 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 367 {
 368         struct gfs2_trans *tr, *safe;
 369         unsigned int old_tail = sdp->sd_log_tail;
 370         int wrap = (new_tail < old_tail);
 371         int a, b, rm;
 372
 373         spin_lock(&sdp->sd_ail_lock);
 374
 375         list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) {
 376                 a = (old_tail <= tr->tr_first);
 377                 b = (tr->tr_first < new_tail);
 378                 rm = (wrap) ? (a || b) : (a && b);
 379                 if (!rm)
 380                         continue;
 381
 382                 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
 383                 list_del(&tr->tr_list);
 384                 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
 385                 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
 386                 gfs2_trans_free(sdp, tr);
 387         }
 388
 389         spin_unlock(&sdp->sd_ail_lock);
 390 }
 391
 392 /**
 393  * gfs2_log_release - Release a given number of log blocks
 394  * @sdp: The GFS2 superblock
 395  * @blks: The number of blocks
 396  *
 397  */
 398
 399 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
 400 {
 401
 402         atomic_add(blks, &sdp->sd_log_blks_free);
 403         trace_gfs2_log_blocks(sdp, blks);
 404         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 405                                   sdp->sd_jdesc->jd_blocks);
 406         up_read(&sdp->sd_log_flush_lock);
 407 }
 408
 409 /**
 410  * gfs2_log_reserve - Make a log reservation
 411  * @sdp: The GFS2 superblock
 412  * @blks: The number of blocks to reserve
 413  *
 414  * Note that we never give out the last few blocks of the journal. Thats
 415  * due to the fact that there is a small number of header blocks
 416  * associated with each log flush. The exact number can't be known until
 417  * flush time, so we ensure that we have just enough free blocks at all
 418  * times to avoid running out during a log flush.
 419  *
 420  * We no longer flush the log here, instead we wake up logd to do that
 421  * for us. To avoid the thundering herd and to ensure that we deal fairly
 422  * with queued waiters, we use an exclusive wait. This means that when we
 423  * get woken with enough journal space to get our reservation, we need to
 424  * wake the next waiter on the list.
 425  *
 426  * Returns: errno
 427  */
 428
 429 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
 430 {
 431         int ret = 0;
 432         unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize);
 433         unsigned wanted = blks + reserved_blks;
 434         DEFINE_WAIT(wait);
 435         int did_wait = 0;
 436         unsigned int free_blocks;
 437
 438         if (gfs2_assert_warn(sdp, blks) ||
 439             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
 440                 return -EINVAL;
 441         atomic_add(blks, &sdp->sd_log_blks_needed);
 442 retry:
 443         free_blocks = atomic_read(&sdp->sd_log_blks_free);
 444         if (unlikely(free_blocks <= wanted)) {
 445                 do {
 446                         prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
 447                                         TASK_UNINTERRUPTIBLE);
 448                         wake_up(&sdp->sd_logd_waitq);
 449                         did_wait = 1;
 450                         if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
 451                                 io_schedule();
 452                         free_blocks = atomic_read(&sdp->sd_log_blks_free);
 453                 } while(free_blocks <= wanted);
 454                 finish_wait(&sdp->sd_log_waitq, &wait);
 455         }
 456         atomic_inc(&sdp->sd_reserving_log);
 457         if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks,
 458                                 free_blocks - blks) != free_blocks) {
 459                 if (atomic_dec_and_test(&sdp->sd_reserving_log))
 460                         wake_up(&sdp->sd_reserving_log_wait);
 461                 goto retry;
 462         }
 463         atomic_sub(blks, &sdp->sd_log_blks_needed);
 464         trace_gfs2_log_blocks(sdp, -blks);
 465
 466         /*
 467          * If we waited, then so might others, wake them up _after_ we get
 468          * our share of the log.
 469          */
 470         if (unlikely(did_wait))
 471                 wake_up(&sdp->sd_log_waitq);
 472
 473         down_read(&sdp->sd_log_flush_lock);
 474         if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
 475                 gfs2_log_release(sdp, blks);
 476                 ret = -EROFS;
 477         }
 478         if (atomic_dec_and_test(&sdp->sd_reserving_log))
 479                 wake_up(&sdp->sd_reserving_log_wait);
 480         return ret;
 481 }
 482
 483 /**
 484  * log_distance - Compute distance between two journal blocks
 485  * @sdp: The GFS2 superblock
 486  * @newer: The most recent journal block of the pair
 487  * @older: The older journal block of the pair
 488  *
 489  *   Compute the distance (in the journal direction) between two
 490  *   blocks in the journal
 491  *
 492  * Returns: the distance in blocks
 493  */
 494
 495 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
 496                                         unsigned int older)
 497 {
 498         int dist;
 499
 500         dist = newer - older;
 501         if (dist < 0)
 502                 dist += sdp->sd_jdesc->jd_blocks;
 503
 504         return dist;
 505 }
 506
 507 /**
 508  * calc_reserved - Calculate the number of blocks to reserve when
 509  *                 refunding a transaction's unused buffers.
 510  * @sdp: The GFS2 superblock
 511  *
 512  * This is complex.  We need to reserve room for all our currently used
 513  * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
 514  * all our journaled data buffers for journaled files (e.g. files in the
 515  * meta_fs like rindex, or files for which chattr +j was done.)
 516  * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
 517  * will count it as free space (sd_log_blks_free) and corruption will follow.
 518  *
 519  * We can have metadata bufs and jdata bufs in the same journal.  So each
 520  * type gets its own log header, for which we need to reserve a block.
 521  * In fact, each type has the potential for needing more than one header
 522  * in cases where we have more buffers than will fit on a journal page.
 523  * Metadata journal entries take up half the space of journaled buffer entries.
 524  * Thus, metadata entries have buf_limit (502) and journaled buffers have
 525  * databuf_limit (251) before they cause a wrap around.
 526  *
 527  * Also, we need to reserve blocks for revoke journal entries and one for an
 528  * overall header for the lot.
 529  *
 530  * Returns: the number of blocks reserved
 531  */
 532 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
 533 {
 534         unsigned int reserved = 0;
 535         unsigned int mbuf;
 536         unsigned int dbuf;
 537         struct gfs2_trans *tr = sdp->sd_log_tr;
 538
 539         if (tr) {
 540                 mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
 541                 dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
 542                 reserved = mbuf + dbuf;
 543                 /* Account for header blocks */
 544                 reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp));
 545                 reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp));
 546         }
 547
 548         if (sdp->sd_log_committed_revoke > 0)
 549                 reserved += gfs2_struct2blk(sdp, sdp->sd_log_committed_revoke);
 550         /* One for the overall header */
 551         if (reserved)
 552                 reserved++;
 553         return reserved;
 554 }
 555
 556 static unsigned int current_tail(struct gfs2_sbd *sdp)
 557 {
 558         struct gfs2_trans *tr;
 559         unsigned int tail;
 560
 561         spin_lock(&sdp->sd_ail_lock);
 562
 563         if (list_empty(&sdp->sd_ail1_list)) {
 564                 tail = sdp->sd_log_head;
 565         } else {
 566                 tr = list_last_entry(&sdp->sd_ail1_list, struct gfs2_trans,
 567                                 tr_list);
 568                 tail = tr->tr_first;
 569         }
 570
 571         spin_unlock(&sdp->sd_ail_lock);
 572
 573         return tail;
 574 }
 575
 576 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
 577 {
 578         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 579
 580         ail2_empty(sdp, new_tail);
 581
 582         atomic_add(dist, &sdp->sd_log_blks_free);
 583         trace_gfs2_log_blocks(sdp, dist);
 584         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 585                              sdp->sd_jdesc->jd_blocks);
 586
 587         sdp->sd_log_tail = new_tail;
 588 }
 589
 590
 591 void log_flush_wait(struct gfs2_sbd *sdp)
 592 {
 593         DEFINE_WAIT(wait);
 594
 595         if (atomic_read(&sdp->sd_log_in_flight)) {
 596                 do {
 597                         prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
 598                                         TASK_UNINTERRUPTIBLE);
 599                         if (atomic_read(&sdp->sd_log_in_flight))
 600                                 io_schedule();
 601                 } while(atomic_read(&sdp->sd_log_in_flight));
 602                 finish_wait(&sdp->sd_log_flush_wait, &wait);
 603         }
 604 }
 605
 606 static int ip_cmp(void *priv, struct list_head *a, struct list_head *b)
 607 {
 608         struct gfs2_inode *ipa, *ipb;
 609
 610         ipa = list_entry(a, struct gfs2_inode, i_ordered);
 611         ipb = list_entry(b, struct gfs2_inode, i_ordered);
 612
 613         if (ipa->i_no_addr < ipb->i_no_addr)
 614                 return -1;
 615         if (ipa->i_no_addr > ipb->i_no_addr)
 616                 return 1;
 617         return 0;
 618 }
 619
 620 static void __ordered_del_inode(struct gfs2_inode *ip)
 621 {
 622         if (!list_empty(&ip->i_ordered))
 623                 list_del_init(&ip->i_ordered);
 624 }
 625
 626 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
 627 {
 628         struct gfs2_inode *ip;
 629         LIST_HEAD(written);
 630
 631         spin_lock(&sdp->sd_ordered_lock);
 632         list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp);
 633         while (!list_empty(&sdp->sd_log_ordered)) {
 634                 ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
 635                 if (ip->i_inode.i_mapping->nrpages == 0) {
 636                         __ordered_del_inode(ip);
 637                         continue;
 638                 }
 639                 list_move(&ip->i_ordered, &written);
 640                 spin_unlock(&sdp->sd_ordered_lock);
 641                 filemap_fdatawrite(ip->i_inode.i_mapping);
 642                 spin_lock(&sdp->sd_ordered_lock);
 643         }
 644         list_splice(&written, &sdp->sd_log_ordered);
 645         spin_unlock(&sdp->sd_ordered_lock);
 646 }
 647
 648 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
 649 {
 650         struct gfs2_inode *ip;
 651
 652         spin_lock(&sdp->sd_ordered_lock);
 653         while (!list_empty(&sdp->sd_log_ordered)) {
 654                 ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
 655                 __ordered_del_inode(ip);
 656                 if (ip->i_inode.i_mapping->nrpages == 0)
 657                         continue;
 658                 spin_unlock(&sdp->sd_ordered_lock);
 659                 filemap_fdatawait(ip->i_inode.i_mapping);
 660                 spin_lock(&sdp->sd_ordered_lock);
 661         }
 662         spin_unlock(&sdp->sd_ordered_lock);
 663 }
 664
 665 void gfs2_ordered_del_inode(struct gfs2_inode *ip)
 666 {
 667         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 668
 669         spin_lock(&sdp->sd_ordered_lock);
 670         __ordered_del_inode(ip);
 671         spin_unlock(&sdp->sd_ordered_lock);
 672 }
 673
 674 void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
 675 {
 676         struct buffer_head *bh = bd->bd_bh;
 677         struct gfs2_glock *gl = bd->bd_gl;
 678
 679         sdp->sd_log_num_revoke++;
 680         if (atomic_inc_return(&gl->gl_revokes) == 1)
 681                 gfs2_glock_hold(gl);
 682         bh->b_private = NULL;
 683         bd->bd_blkno = bh->b_blocknr;
 684         gfs2_remove_from_ail(bd); /* drops ref on bh */
 685         bd->bd_bh = NULL;
 686         set_bit(GLF_LFLUSH, &gl->gl_flags);
 687         list_add(&bd->bd_list, &sdp->sd_log_revokes);
 688 }
 689
 690 void gfs2_glock_remove_revoke(struct gfs2_glock *gl)
 691 {
 692         if (atomic_dec_return(&gl->gl_revokes) == 0) {
 693                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 694                 gfs2_glock_queue_put(gl);
 695         }
 696 }
 697
 698 /**
 699  * gfs2_write_revokes - Add as many revokes to the system transaction as we can
 700  * @sdp: The GFS2 superblock
 701  *
 702  * Our usual strategy is to defer writing revokes as much as we can in the hope
 703  * that we'll eventually overwrite the journal, which will make those revokes
 704  * go away.  This changes when we flush the log: at that point, there will
 705  * likely be some left-over space in the last revoke block of that transaction.
 706  * We can fill that space with additional revokes for blocks that have already
 707  * been written back.  This will basically come at no cost now, and will save
 708  * us from having to keep track of those blocks on the AIL2 list later.
 709  */
 710 void gfs2_write_revokes(struct gfs2_sbd *sdp)
 711 {
 712         /* number of revokes we still have room for */
 713         int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
 714
 715         gfs2_log_lock(sdp);
 716         while (sdp->sd_log_num_revoke > max_revokes)
 717                 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
 718         max_revokes -= sdp->sd_log_num_revoke;
 719         if (!sdp->sd_log_num_revoke) {
 720                 atomic_dec(&sdp->sd_log_blks_free);
 721                 /* If no blocks have been reserved, we need to also
 722                  * reserve a block for the header */
 723                 if (!sdp->sd_log_blks_reserved) {
 724                         atomic_dec(&sdp->sd_log_blks_free);
 725                         trace_gfs2_log_blocks(sdp, -2);
 726                 } else {
 727                         trace_gfs2_log_blocks(sdp, -1);
 728                 }
 729         }
 730         gfs2_ail1_empty(sdp, max_revokes);
 731         gfs2_log_unlock(sdp);
 732
 733         if (!sdp->sd_log_num_revoke) {
 734                 atomic_inc(&sdp->sd_log_blks_free);
 735                 if (!sdp->sd_log_blks_reserved) {
 736                         atomic_inc(&sdp->sd_log_blks_free);
 737                         trace_gfs2_log_blocks(sdp, 2);
 738                 } else {
 739                         trace_gfs2_log_blocks(sdp, 1);
 740                 }
 741         }
 742 }
 743
 744 /**
 745  * gfs2_write_log_header - Write a journal log header buffer at lblock
 746  * @sdp: The GFS2 superblock
 747  * @jd: journal descriptor of the journal to which we are writing
 748  * @seq: sequence number
 749  * @tail: tail of the log
 750  * @lblock: value for lh_blkno (block number relative to start of journal)
 751  * @flags: log header flags GFS2_LOG_HEAD_*
 752  * @op_flags: flags to pass to the bio
 753  *
 754  * Returns: the initialized log buffer descriptor
 755  */
 756
 757 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
 758                            u64 seq, u32 tail, u32 lblock, u32 flags,
 759                            int op_flags)
 760 {
 761         struct gfs2_log_header *lh;
 762         u32 hash, crc;
 763         struct page *page;
 764         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 765         struct timespec64 tv;
 766         struct super_block *sb = sdp->sd_vfs;
 767         u64 dblock;
 768
 769         if (gfs2_withdrawn(sdp))
 770                 goto out;
 771
 772         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 773         lh = page_address(page);
 774         clear_page(lh);
 775
 776         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 777         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 778         lh->lh_header.__pad0 = cpu_to_be64(0);
 779         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 780         lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
 781         lh->lh_sequence = cpu_to_be64(seq);
 782         lh->lh_flags = cpu_to_be32(flags);
 783         lh->lh_tail = cpu_to_be32(tail);
 784         lh->lh_blkno = cpu_to_be32(lblock);
 785         hash = ~crc32(~0, lh, LH_V1_SIZE);
 786         lh->lh_hash = cpu_to_be32(hash);
 787
 788         ktime_get_coarse_real_ts64(&tv);
 789         lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
 790         lh->lh_sec = cpu_to_be64(tv.tv_sec);
 791         if (!list_empty(&jd->extent_list))
 792                 dblock = gfs2_log_bmap(jd, lblock);
 793         else {
 794                 int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock);
 795                 if (gfs2_assert_withdraw(sdp, ret == 0))
 796                         return;
 797         }
 798         lh->lh_addr = cpu_to_be64(dblock);
 799         lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
 800
 801         /* We may only write local statfs, quota, etc., when writing to our
 802            own journal. The values are left 0 when recovering a journal
 803            different from our own. */
 804         if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
 805                 lh->lh_statfs_addr =
 806                         cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
 807                 lh->lh_quota_addr =
 808                         cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
 809
 810                 spin_lock(&sdp->sd_statfs_spin);
 811                 lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
 812                 lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
 813                 lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
 814                 spin_unlock(&sdp->sd_statfs_spin);
 815         }
 816
 817         BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
 818
 819         crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
 820                      sb->s_blocksize - LH_V1_SIZE - 4);
 821         lh->lh_crc = cpu_to_be32(crc);
 822
 823         gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock);
 824         gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags);
 825 out:
 826         log_flush_wait(sdp);
 827 }
 828
 829 /**
 830  * log_write_header - Get and initialize a journal header buffer
 831  * @sdp: The GFS2 superblock
 832  * @flags: The log header flags, including log header origin
 833  *
 834  * Returns: the initialized log buffer descriptor
 835  */
 836
 837 static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
 838 {
 839         unsigned int tail;
 840         int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
 841         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 842
 843         gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
 844         tail = current_tail(sdp);
 845
 846         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
 847                 gfs2_ordered_wait(sdp);
 848                 log_flush_wait(sdp);
 849                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
 850         }
 851         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 852         gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail,
 853                               sdp->sd_log_flush_head, flags, op_flags);
 854         gfs2_log_incr_head(sdp);
 855
 856         if (sdp->sd_log_tail != tail)
 857                 log_pull_tail(sdp, tail);
 858 }
 859
 860 /**
 861  * ail_drain - drain the ail lists after a withdraw
 862  * @sdp: Pointer to GFS2 superblock
 863  */
 864 static void ail_drain(struct gfs2_sbd *sdp)
 865 {
 866         struct gfs2_trans *tr;
 867
 868         spin_lock(&sdp->sd_ail_lock);
 869         /*
 870          * For transactions on the sd_ail1_list we need to drain both the
 871          * ail1 and ail2 lists. That's because function gfs2_ail1_start_one
 872          * (temporarily) moves items from its tr_ail1 list to tr_ail2 list
 873          * before revokes are sent for that block. Items on the sd_ail2_list
 874          * should have already gotten beyond that point, so no need.
 875          */
 876         while (!list_empty(&sdp->sd_ail1_list)) {
 877                 tr = list_first_entry(&sdp->sd_ail1_list, struct gfs2_trans,
 878                                       tr_list);
 879                 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail1_list);
 880                 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
 881                 list_del(&tr->tr_list);
 882                 gfs2_trans_free(sdp, tr);
 883         }
 884         while (!list_empty(&sdp->sd_ail2_list)) {
 885                 tr = list_first_entry(&sdp->sd_ail2_list, struct gfs2_trans,
 886                                       tr_list);
 887                 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
 888                 list_del(&tr->tr_list);
 889                 gfs2_trans_free(sdp, tr);
 890         }
 891         spin_unlock(&sdp->sd_ail_lock);
 892 }
 893
 894 /**
 895  * empty_ail1_list - try to start IO and empty the ail1 list
 896  * @sdp: Pointer to GFS2 superblock
 897  */
 898 static void empty_ail1_list(struct gfs2_sbd *sdp)
 899 {
 900         unsigned long start = jiffies;
 901
 902         for (;;) {
 903                 if (time_after(jiffies, start + (HZ * 600))) {
 904                         fs_err(sdp, "Error: In %s for 10 minutes! t=%d\n",
 905                                __func__, current->journal_info ? 1 : 0);
 906                         dump_ail_list(sdp);
 907                         return;
 908                 }
 909                 gfs2_ail1_start(sdp);
 910                 gfs2_ail1_wait(sdp);
 911                 if (gfs2_ail1_empty(sdp, 0))
 912                         return;
 913         }
 914 }
 915
 916 /**
 917  * trans_drain - drain the buf and databuf queue for a failed transaction
 918  * @tr: the transaction to drain
 919  *
 920  * When this is called, we're taking an error exit for a log write that failed
 921  * but since we bypassed the after_commit functions, we need to remove the
 922  * items from the buf and databuf queue.
 923  */
 924 static void trans_drain(struct gfs2_trans *tr)
 925 {
 926         struct gfs2_bufdata *bd;
 927         struct list_head *head;
 928
 929         if (!tr)
 930                 return;
 931
 932         head = &tr->tr_buf;
 933         while (!list_empty(head)) {
 934                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 935                 list_del_init(&bd->bd_list);
 936                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 937         }
 938         head = &tr->tr_databuf;
 939         while (!list_empty(head)) {
 940                 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
 941                 list_del_init(&bd->bd_list);
 942                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 943         }
 944 }
 945
 946 /**
 947  * gfs2_log_flush - flush incore transaction(s)
 948  * @sdp: the filesystem
 949  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 950  * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
 951  *
 952  */
 953
 954 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
 955 {
 956         struct gfs2_trans *tr = NULL;
 957         enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
 958
 959         down_write(&sdp->sd_log_flush_lock);
 960
 961         /*
 962          * Do this check while holding the log_flush_lock to prevent new
 963          * buffers from being added to the ail via gfs2_pin()
 964          */
 965         if (gfs2_withdrawn(sdp))
 966                 goto out;
 967
 968         /* Log might have been flushed while we waited for the flush lock */
 969         if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags))
 970                 goto out;
 971         trace_gfs2_log_flush(sdp, 1, flags);
 972
 973         if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
 974                 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 975
 976         sdp->sd_log_flush_head = sdp->sd_log_head;
 977         tr = sdp->sd_log_tr;
 978         if (tr) {
 979                 sdp->sd_log_tr = NULL;
 980                 tr->tr_first = sdp->sd_log_flush_head;
 981                 if (unlikely (state == SFS_FROZEN))
 982                         if (gfs2_assert_withdraw_delayed(sdp,
 983                                !tr->tr_num_buf_new && !tr->tr_num_databuf_new))
 984                                 goto out_withdraw;
 985         }
 986
 987         if (unlikely(state == SFS_FROZEN))
 988                 if (gfs2_assert_withdraw_delayed(sdp, !sdp->sd_log_num_revoke))
 989                         goto out_withdraw;
 990         if (gfs2_assert_withdraw_delayed(sdp,
 991                         sdp->sd_log_num_revoke == sdp->sd_log_committed_revoke))
 992                 goto out_withdraw;
 993
 994         gfs2_ordered_write(sdp);
 995         if (gfs2_withdrawn(sdp))
 996                 goto out_withdraw;
 997         lops_before_commit(sdp, tr);
 998         if (gfs2_withdrawn(sdp))
 999                 goto out_withdraw;
1000         gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE);
1001         if (gfs2_withdrawn(sdp))
1002                 goto out_withdraw;
1003
1004         if (sdp->sd_log_head != sdp->sd_log_flush_head) {
1005                 log_flush_wait(sdp);
1006                 log_write_header(sdp, flags);
1007         } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
1008                 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
1009                 trace_gfs2_log_blocks(sdp, -1);
1010                 log_write_header(sdp, flags);
1011         }
1012         if (gfs2_withdrawn(sdp))
1013                 goto out_withdraw;
1014         lops_after_commit(sdp, tr);
1015
1016         gfs2_log_lock(sdp);
1017         sdp->sd_log_head = sdp->sd_log_flush_head;
1018         sdp->sd_log_blks_reserved = 0;
1019         sdp->sd_log_committed_revoke = 0;
1020
1021         spin_lock(&sdp->sd_ail_lock);
1022         if (tr && !list_empty(&tr->tr_ail1_list)) {
1023                 list_add(&tr->tr_list, &sdp->sd_ail1_list);
1024                 tr = NULL;
1025         }
1026         spin_unlock(&sdp->sd_ail_lock);
1027         gfs2_log_unlock(sdp);
1028
1029         if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
1030                 if (!sdp->sd_log_idle) {
1031                         empty_ail1_list(sdp);
1032                         if (gfs2_withdrawn(sdp))
1033                                 goto out_withdraw;
1034                         atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
1035                         trace_gfs2_log_blocks(sdp, -1);
1036                         log_write_header(sdp, flags);
1037                         sdp->sd_log_head = sdp->sd_log_flush_head;
1038                 }
1039                 if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
1040                              GFS2_LOG_HEAD_FLUSH_FREEZE))
1041                         gfs2_log_shutdown(sdp);
1042                 if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
1043                         atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
1044         }
1045
1046 out_end:
1047         trace_gfs2_log_flush(sdp, 0, flags);
1048 out:
1049         up_write(&sdp->sd_log_flush_lock);
1050         gfs2_trans_free(sdp, tr);
1051         if (gfs2_withdrawing(sdp))
1052                 gfs2_withdraw(sdp);
1053         return;
1054
1055 out_withdraw:
1056         trans_drain(tr);
1057         /**
1058          * If the tr_list is empty, we're withdrawing during a log
1059          * flush that targets a transaction, but the transaction was
1060          * never queued onto any of the ail lists. Here we add it to
1061          * ail1 just so that ail_drain() will find and free it.
1062          */
1063         spin_lock(&sdp->sd_ail_lock);
1064         if (tr && list_empty(&tr->tr_list))
1065                 list_add(&tr->tr_list, &sdp->sd_ail1_list);
1066         spin_unlock(&sdp->sd_ail_lock);
1067         ail_drain(sdp); /* frees all transactions */
1068         tr = NULL;
1069         goto out_end;
1070 }
1071
1072 /**
1073  * gfs2_merge_trans - Merge a new transaction into a cached transaction
1074  * @old: Original transaction to be expanded
1075  * @new: New transaction to be merged
1076  */
1077
1078 static void gfs2_merge_trans(struct gfs2_sbd *sdp, struct gfs2_trans *new)
1079 {
1080         struct gfs2_trans *old = sdp->sd_log_tr;
1081
1082         WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
1083
1084         old->tr_num_buf_new     += new->tr_num_buf_new;
1085         old->tr_num_databuf_new += new->tr_num_databuf_new;
1086         old->tr_num_buf_rm      += new->tr_num_buf_rm;
1087         old->tr_num_databuf_rm  += new->tr_num_databuf_rm;
1088         old->tr_num_revoke      += new->tr_num_revoke;
1089         old->tr_num_revoke_rm   += new->tr_num_revoke_rm;
1090
1091         list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
1092         list_splice_tail_init(&new->tr_buf, &old->tr_buf);
1093
1094         spin_lock(&sdp->sd_ail_lock);
1095         list_splice_tail_init(&new->tr_ail1_list, &old->tr_ail1_list);
1096         list_splice_tail_init(&new->tr_ail2_list, &old->tr_ail2_list);
1097         spin_unlock(&sdp->sd_ail_lock);
1098 }
1099
1100 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1101 {
1102         unsigned int reserved;
1103         unsigned int unused;
1104         unsigned int maxres;
1105
1106         gfs2_log_lock(sdp);
1107
1108         if (sdp->sd_log_tr) {
1109                 gfs2_merge_trans(sdp, tr);
1110         } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
1111                 gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
1112                 sdp->sd_log_tr = tr;
1113                 set_bit(TR_ATTACHED, &tr->tr_flags);
1114         }
1115
1116         sdp->sd_log_committed_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
1117         reserved = calc_reserved(sdp);
1118         maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
1119         gfs2_assert_withdraw(sdp, maxres >= reserved);
1120         unused = maxres - reserved;
1121         atomic_add(unused, &sdp->sd_log_blks_free);
1122         trace_gfs2_log_blocks(sdp, unused);
1123         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
1124                              sdp->sd_jdesc->jd_blocks);
1125         sdp->sd_log_blks_reserved = reserved;
1126
1127         gfs2_log_unlock(sdp);
1128 }
1129
1130 /**
1131  * gfs2_log_commit - Commit a transaction to the log
1132  * @sdp: the filesystem
1133  * @tr: the transaction
1134  *
1135  * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
1136  * or the total number of used blocks (pinned blocks plus AIL blocks)
1137  * is greater than thresh2.
1138  *
1139  * At mount time thresh1 is 2/5ths of journal size, thresh2 is 4/5ths of
1140  * journal size.
1141  *
1142  * Returns: errno
1143  */
1144
1145 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1146 {
1147         log_refund(sdp, tr);
1148
1149         if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
1150             ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
1151             atomic_read(&sdp->sd_log_thresh2)))
1152                 wake_up(&sdp->sd_logd_waitq);
1153 }
1154
1155 /**
1156  * gfs2_log_shutdown - write a shutdown header into a journal
1157  * @sdp: the filesystem
1158  *
1159  */
1160
1161 static void gfs2_log_shutdown(struct gfs2_sbd *sdp)
1162 {
1163         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
1164         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
1165         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
1166
1167         sdp->sd_log_flush_head = sdp->sd_log_head;
1168
1169         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
1170
1171         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
1172         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
1173
1174         sdp->sd_log_head = sdp->sd_log_flush_head;
1175         sdp->sd_log_tail = sdp->sd_log_head;
1176 }
1177
1178 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
1179 {
1180         return (atomic_read(&sdp->sd_log_pinned) +
1181                 atomic_read(&sdp->sd_log_blks_needed) >=
1182                 atomic_read(&sdp->sd_log_thresh1));
1183 }
1184
1185 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
1186 {
1187         unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
1188
1189         if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
1190                 return 1;
1191
1192         return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
1193                 atomic_read(&sdp->sd_log_thresh2);
1194 }
1195
1196 /**
1197  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
1198  * @sdp: Pointer to GFS2 superblock
1199  *
1200  * Also, periodically check to make sure that we're using the most recent
1201  * journal index.
1202  */
1203
1204 int gfs2_logd(void *data)
1205 {
1206         struct gfs2_sbd *sdp = data;
1207         unsigned long t = 1;
1208         DEFINE_WAIT(wait);
1209         bool did_flush;
1210
1211         while (!kthread_should_stop()) {
1212
1213                 if (gfs2_withdrawn(sdp)) {
1214                         msleep_interruptible(HZ);
1215                         continue;
1216                 }
1217                 /* Check for errors writing to the journal */
1218                 if (sdp->sd_log_error) {
1219                         gfs2_lm(sdp,
1220                                 "GFS2: fsid=%s: error %d: "
1221                                 "withdrawing the file system to "
1222                                 "prevent further damage.\n",
1223                                 sdp->sd_fsname, sdp->sd_log_error);
1224                         gfs2_withdraw(sdp);
1225                         continue;
1226                 }
1227
1228                 did_flush = false;
1229                 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1230                         gfs2_ail1_empty(sdp, 0);
1231                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1232                                        GFS2_LFC_LOGD_JFLUSH_REQD);
1233                         did_flush = true;
1234                 }
1235
1236                 if (gfs2_ail_flush_reqd(sdp)) {
1237                         gfs2_ail1_start(sdp);
1238                         gfs2_ail1_wait(sdp);
1239                         gfs2_ail1_empty(sdp, 0);
1240                         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1241                                        GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1242                         did_flush = true;
1243                 }
1244
1245                 if (!gfs2_ail_flush_reqd(sdp) || did_flush)
1246                         wake_up(&sdp->sd_log_waitq);
1247
1248                 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1249
1250                 try_to_freeze();
1251
1252                 do {
1253                         prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1254                                         TASK_INTERRUPTIBLE);
1255                         if (!gfs2_ail_flush_reqd(sdp) &&
1256                             !gfs2_jrnl_flush_reqd(sdp) &&
1257                             !kthread_should_stop())
1258                                 t = schedule_timeout(t);
1259                 } while(t && !gfs2_ail_flush_reqd(sdp) &&
1260                         !gfs2_jrnl_flush_reqd(sdp) &&
1261                         !kthread_should_stop());
1262                 finish_wait(&sdp->sd_logd_waitq, &wait);
1263         }
1264
1265         return 0;
1266 }
1267